ivy.el (ivy-immediate-done): Exit with unchanged initial input

[nxtr]

This change makes ivy-immediate-done (without any editing) to exit
completing-read with return value in same (abbreviated) form and value
as the initial input.

This fix functions read-file-name and read-directory-name to return the
default-filename with ivy-immediate-done, instead of current
input (expanded).

Fixes #1170

[abo-abo]

Thanks, the changes look good. However, they are together over 15 lines. Do you have an Emacs CA?

[nxtr]

Great! Not yet, been waiting 7+ days for the form. I'll get back once signed and posted.

[abo-abo]

Thanks.

[nxtr]

Finally, the form is signed, posted and returned.

[basil-conto]

Nitpick: if default-filename is referring to the eponymous read-file-name argument, then I suggest formatting it as DEFAULT-FILENAME, as is already done with INITIAL-INPUT.

[nxtr]

Not quite. IICR, default-filename refers to the (setq default-filename ... ) value in read-file-name-default. Still, good point and the comment wording should be edited and rephrased.

[basil-conto]

Not quite. IICR, default-filename refers to the (setq default-filename ... ) value in read-file-name-default.

Okay, but that local variable still corresponds to its enclosing function's positional argument of the same name, no?

Still, good point and the comment wording should be edited and rephrased.

That you added a comment at all is already very useful, but I am always grateful for efforts in improving documentation, thanks!

[nxtr]

Okay, but that local variable still corresponds to its enclosing function's positional argument of the same name, no?

Yup, here the focus was more on the value to be returned, not necessarily the provided value, that's all. I have no problem with an upcase.

[basil-conto]

Yup, here the focus was more on the value to be returned, not necessarily the provided value, that's all.

Gotcha. In that case I suggest explicitly referring to the function's return value, rather than to some implementation detail.

[basil-conto]

Nitpick: I wonder if a cond would avoid the need for nested conditionals?

(setf (ivy-state-current ivy-last)
      (cond ((or (not ivy--directory)
                 (eq (ivy-state-history ivy-last) 'grep-files-history))
             ivy-text)
            ((and (string= ivy-text "")
                  (eq (ivy-state-collection ivy-last)
                      #'read-file-name-internal))
             ;; For `read-file-name' compat, unchanged initial input means
             ;; that `ivy-read' shall return INITIAL-INPUT.
             ;; `read-file-name-default' `string-equal' return value with
             ;; provided INITIAL-INPUT to detect that the user choose the
             ;; default, DEFAULT-FILENAME.  We must return `ivy--directory'
             ;; in unchanged form in cased `ivy--directory' started out as
             ;; INITIAL-INPUT in abbreviated form.
             ivy--directory)          ; Unchanged (unexpanded)
            (t
             (expand-file-name ivy-text ivy--directory)))))
(insert (ivy-state-current ivy-last))

[nxtr]

Yup, nice, was more into adding than rewriting at the time.

[basil-conto]

I usually collect minor refactors over time, so I'll just add this to my pile, unless someone beats me to it.

[abo-abo]

Thanks.